Samsung Introduces SOCAMM2, a Detachable LPDDR5X Server Memory Module Aimed at AI Infrastructure

Posted on Edited on
Samsung introduced SOCAMM2, a detachable LPDDR5X-based server memory module for AI systems, claiming over 2x RDIMM bandwidth and more than 55% lower power.

Quick Report

Samsung has announced SOCAMM2 (Small Outline Compression Attached Memory Module), a server-focused, LPDDR-based memory module designed for AI data center workloads. The company says SOCAMM2 combines LPDDR power characteristics with a modular, detachable form factor, and claims it can deliver more than twice the bandwidth of traditional RDIMMs while using over 55% less power.

A key architectural selling point is serviceability: unlike soldered LPDDR designs, SOCAMM2 is positioned as upgradeable and replaceable, potentially reducing downtime and lowering total cost of ownership over a server's lifecycle.

Samsung also frames SOCAMM2 as a mechanical and thermal design improvement. The press release argues that shifting away from a vertical RDIMM layout to a more horizontal module orientation can improve space utilization and make heatsink and airflow design easier, including compatibility with air and liquid cooling.

On ecosystem support, Samsung says it is collaborating with NVIDIA to optimize SOCAMM2 for NVIDIA accelerated infrastructure, and that JEDEC standardization work is underway for LPDDR-based server modules.

Written using GitHub Copilot GPT-5.2 in agentic mode instructed to follow current codebase style and conventions for writing articles.

Source(s)

  • TPU
  • Samsung (press release)